An alternating least squares algorithm for fitting the two- and three-way dedicom model and the idioscal model

The DEDICOM model is a model for representing asymmetric relations among a set of objects by means of a set of coordinates for the objects on a limited number of dimensions. The present paper offers an alternating least squares algorithm for fitting the DEDICOM model. The model can be generalized to represent any number of sets of relations among the same set of objects. An algorithm for fitting this three-way DEDICOM model is provided as well. Based on the algorithm for the three-way DEDICOM model an algorithm is developed for fitting the IDIOSCAL model in the least squares sense.The author is obliged to Jos ten Berge and Richard Harshman.     

A generalization of Takane's algorithm for dedicom

An algorithm is described for fitting the DEDICOM model for the analysis of asymmetric data matrices. This algorithm generalizes an algorithm suggested by Takane in that it uses a damping parameter in the iterative process. Takane's algorithm does not always converge monotonically. Based on the generalized algorithm, a modification of Takane's algorithm is suggested such that this modified algorithm converges monotonically. It is suggested to choose as starting configurations for the algorithm those configurations that yield closed-form solutions in some special cases. Finally, a sufficient condition is described for monotonic convergence of Takane's original algorithm.Financial Support by the Netherlands organization for scientific research (NWO) is gratefully acknowledged. The authors are obliged to Richard Harshman.     

The multidimensional analysis of asymmetries in alphabetic confusion matrices: Evidence for global-to-local and local-to-global processing

This study examined the ability of an asymmetric multidimensional scaling program (DEDICOM) to reveal information about letter-perception processes. To demonstrate its potential, we applied it to the controversy concerning local-to-global versus global-to-local letter perception. These two theories lead to different predictions about stimulus confusion asymmetries. Since DEDICOM is capable of recovering the structure of asymmetric or directional patterns, it should reveal whether a stimulus-response confusion matrix contains patterns of asymmetry more consistent with one or the other perceptual theory. This was tested using two data sets. The first (from Lupker, 1979) revealed an additive hierarchy of asymmetry strongly consistent with global-tolocal processing, although unexpected additional structure and reliable anomalies indicated the need for a more refined theoretical account. The second (a full alphabetic confusion matrix combining data from Gilmore et al., 1979; Loomis, 1982; and Towasend, 1971) revealed five distinct patterns, each consisting of transformations attributable to the failure to detect specific local letter features. This solution strengthened support for local-to-global processing, in sharp contrast to the first analysis. Possible reasons for this divergence are discussed, including differences in the stimuli, exposure durations, and a hypothetical two-stage process of perception. Despite their differences, both solutions demonstrated how asymmetric scaling can reveal structure in asymmetries, which is relevant to perceptual theory and which would have been difficult to recover by other means.     

Integer programs for one- and two-mode blockmodeling based on prespecified image matrices for structural and regular equivalence

Establishing blockmodels for one- and two-mode binary network matrices has typically been accomplished using multiple restarts of heuristic algorithms that minimize functions of inconsistency with an ideal block structure. Although these algorithms likely yield exceptional performance, they are not assured to provide blockmodels that optimize the functional indices. In this paper, we present integer programming models that, for a prespecified image matrix, can produce guaranteed optimal solutions for matrices of nontrivial size. Accordingly, analysts performing a confirmatory analysis of a prespecified blockmodel structure can apply our models directly to obtain an optimal solution. In exploratory cases where a blockmodel structure is not prespecified, we recommend a two-stage procedure, where a heuristic method is first used to identify an image matrix and the integer program is subsequently formulated and solved to identify the optimal solution for that image matrix. Although best suited for ideal block structures associated with structural equivalence, the integer programming models have the flexibility to accommodate functional indices pertaining to regular equivalence. Computational results are reported for a variety of one- and two-mode matrices from the blockmodeling literature.     

On the Non-Existence of Optimal Solutions and the Occurrence of “Degeneracy” in the CANDECOMP/PARAFAC Model

The CANDECOMP/PARAFAC (CP) model decomposes a three-way array into a prespecified number of R factors and a residual array by minimizing the sum of squares of the latter. It is well known that an optimal solution for CP need not exist. We show that if an optimal CP solution does not exist, then any sequence of CP factors monotonically decreasing the CP criterion value to its infimum will exhibit the features of a so-called “degeneracy”. That is, the parameter matrices become nearly rank deficient and the Euclidean norm of some factors tends to infinity. We also show that the CP criterion function does attain its infimum if one of the parameter matrices is constrained to be column-wise orthonormal.     

Constrained DEDICOM

The DEDICOM method for the analysis of asymmetric data tables gives representations that are identified only up to a nonsingular transformation. To identify solutions it is proposed to impose subspace constraints on the stimulus coefficients. Most attention is paid to the case where different subspace constraints are imposed on different dimensions. The procedures are discussed both for the case where the complete table is fitted, and for cases where only offdiagonal elements are fitted. The case where the data table is skew-symmetric is treated separately as well.The research of H. A. L. Kiers has been made possible by a fellowship of the Royal Netherlands Academy of Arts and Sciences. The research of Y. Takane has been supported by the Natural Sciences and Engineering Research Council of Canada, grant number A6394, and by the McGill-IBM Cooperative Grant. The authors are obliged to Richard A. Harshman for helpful comments on an earlier version.     

Latent roots of random data correlation matrices with squared multiple correlations on the diagonal: A monte carlo study

In order to make the parallel analysis criterion for determining the number of factors easy to use, regression equations for predicting the logarithms of the latent roots of random correlation matrices, with squared multiple correlations on the diagonal, are presented. The correlation matrices were derived from distributions of normally distributed random numbers. The independent variables are log (N–1) and log {[n(n–1)/2]–[(i–1)n]}, whereN is the number of observations;n, the number of variables; andi, the ordinal position of the eigenvalue. The results were excellent, with multiple correlation coefficients ranging from .9948 to .9992.This research was supported by the Office of Naval Research under Contract N00014-67-A-0305-0012, Lloyd G. Humphreys, principal investigator, and by the Department of Computer Science of which Richard G. Montanelli, Jr., is a member.     

Graph coloring, minimum-diameter partitioning, and the analysis of confusion matrices

It is well known that minimum-diameter partitioning of symmetric dissimilarity matrices can be framed within the context of coloring the vertices of a graph. Although confusion data are typically represented in the form of asymmetric similarity matrices, they are also amenable to a graph-coloring perspective. In this paper, we propose the integration of the minimum-diameter partitioning method with a neighborhood-based coloring approach for analyzing digraphs corresponding to confusion data. This procedure is capable of producing minimum-diameter partitions with the added desirable property that vertices with the same color have similar in-neighborhoods (i.e., directed edges entering the vertex) and out-neighborhoods (i.e., directed edges exiting the vertex) for the digraph corresponding to the minimum partition diameter.     

Rotational equivalence of factor loading matrices with specified values

Under mild assumptions, when appropriate elements of a factor loading matrix are specified to be zero, all orthogonally equivalent matrices differ at most by column sign changes. Here a variety of results are given for the more complex case when the specified values are not necessarily zero. A method is given for constructing reflections to preserve specified rows and columns. When the appropriatek(k – 1)/2 elements have been specified, sufficient conditions are stated for the existence of 2 ^k orthogonally equivalent matrices.This research was supported in part by the National Institute of Health Grant RR-3.     

Heuristic Implementation of Dynamic Programming for Matrix Permutation Problems in Combinatorial Data Analysis

Dynamic programming methods for matrix permutation problems in combinatorial data analysis can produce globally-optimal solutions for matrices up to size 30×30, but are computationally infeasible for larger matrices because of enormous computer memory requirements. Branch-and-bound methods also guarantee globally-optimal solutions, but computation time considerations generally limit their applicability to matrix sizes no greater than 35×35. Accordingly, a variety of heuristic methods have been proposed for larger matrices, including iterative quadratic assignment, tabu search, simulated annealing, and variable neighborhood search. Although these heuristics can produce exceptional results, they are prone to converge to local optima where the permutation is difficult to dislodge via traditional neighborhood moves (e.g., pairwise interchanges, object-block relocations, object-block reversals, etc.). We show that a heuristic implementation of dynamic programming yields an efficient procedure for escaping local optima. Specifically, we propose applying dynamic programming to reasonably-sized subsequences of consecutive objects in the locally-optimal permutation, identified by simulated annealing, to further improve the value of the objective function. Experimental results are provided for three classic matrix permutation problems in the combinatorial data analysis literature: (a) maximizing a dominance index for an asymmetric proximity matrix; (b) least-squares unidimensional scaling of a symmetric dissimilarity matrix; and (c) approximating an anti-Robinson structure for a symmetric dissimilarity matrix. We are extremely grateful to the Associate Editor and two anonymous reviewers for helpful suggestions and corrections.     

Testing the Domain Specificity of Creativity with Kaufman Domains of Creativity Scale: A Meta-Analytic Confirmatory Factor Analysis

Domain-specificity is a topic of debate within the field of creativity. To shed light on this issue, we conducted a meta-analysis of cross-domain correlations based on the Kaufman Domains of Creativity Scale (K-DOCS). To evaluate the model fit of one general factor versus two factors that encompass the primary K-DOCS subscales (Scholarly, Everyday, Artistic, Scientific, and Performance), we employed the one-stage meta-analytic structural equation modeling (OSMASEM) approach. Poor fit of these models would provide evidence of domain-specificity, as the proposed models would not outperform the independence model. Our analysis included 45 correlation matrices from 30 studies, with a total sample size of 31,136 participants. The results provided support for a general domain of creativity, as well as a two-factor solution consisting of Arts and Sciences factors. Among the subscales, the highest correlation was found between the Artistic and Performance domains (r = .478), while the smallest correlation was observed between the Everyday and Scientific domains (r = .178). Furthermore, moderator analyses incorporating age and gender revealed that the Scientific and Everyday subscales exhibited a stronger factor load in older participants compared to younger participants. Implications are discussed for research and practice.     

Factor analysis for clustered observations

Classical factor analysis assumes a random sample of vectors of observations. For clustered vectors of observations, such as data for students from colleges, or individuals within households, it may be necessary to consider different within-group and between-group factor structures. Such a two-level model for factor analysis is defined, and formulas for a scoring algorithm for estimation with this model are derived. A simple noniterative method based on a decomposition of the total sums of squares and crossproducts is discussed. This method provides a suitable starting solution for the iterative algorithm, but it is also a very good approximation to the maximum likelihood solution. Extensions for higher levels of nesting are indicated. With judicious application of quasi-Newton methods, the amount of computation involved in the scoring algorithm is moderate even for complex problems; in particular, no inversion of matrices with large dimensions is involved. The methods are illustrated on two examples.Suggestions and corrections of three anonymous referees and of an Associate Editor are acknowledged. Discussions with Bob Jennrich on computational aspects were very helpful. Most of research leading to this paper was carried out while the first author was a visiting associate professor at the University of California, Los Angeles.     

Factor Analysis with EM Algorithm Never Gives Improper Solutions when Sample Covariance and Initial Parameter Matrices Are Proper

Rubin and Thayer (Psychometrika, 47:69–76, 1982) proposed the EM algorithm for exploratory and confirmatory maximum likelihood factor analysis. In this paper, we prove the following fact: the EM algorithm always gives a proper solution with positive unique variances and factor correlations with absolute values that do not exceed one, when the covariance matrix to be analyzed and the initial matrices including unique variances and inter-factor correlations are positive definite. We further numerically demonstrate that the EM algorithm yields proper solutions for the data which lead the prevailing gradient algorithms for factor analysis to produce improper solutions. The numerical studies also show that, in real computations with limited numerical precision, Rubin and Thayer’s (Psychometrika, 47:69–76, 1982) original formulas for confirmatory factor analysis can make factor correlation matrices asymmetric, so that the EM algorithm fails to converge. However, this problem can be overcome by using an EM algorithm in which the original formulas are replaced by those guaranteeing the symmetry of factor correlation matrices, or by formulas used to prove the above fact.     

Confirmatory factor analysis of the WPPSI for language-impaired children

The present study used a maximum-likelihood confirmatory factor analysis (CFA) to test the hypothesis that a four-factor model is the most parsimonious explanation of the structure of the WPPSI for language-impaired children. Four separate maximum-likelihood confirmatory factor analyses were performed on a sample of 198 Norwegian language-impaired children tested with the WPPSI, and on the Norwegian (n = 563) and American (n = 1200) standardization samples. A one-factor (general), a two-factor (verbal and performance), three-factor (parallel to the WISC-R) and a four-factor solution composed of "processing dependent" ("knowing how" and "seeing how") and "knowledge dependent" ("knowing that" and "seeing that") were imposed on the average intercorrelation matrices of the 11 WPPSI subtests. Confirmatory factor analysis indicated that the four-factor latent construct model was the most parsimonious explanation of the structure of the WPPSI for language-impaired children, as well as for normally developing children.     

Some uniqueness results for PARAFAC2

Whereas the unique axes properties of PARAFAC1 have been examined extensively, little is known about uniqueness properties for the PARAFAC2 model for covariance matrices. This paper is concerned with uniqueness in the rank two case of PARAFAC2. For this case, Harshman and Lundy have recently shown, subject to mild assumptions, that PARAFAC2 is unique when five (covariance) matrices are analyzed. In the present paper, this result is sharpened. PARAFAC2 is shown to be usually unique with four matrices. With three matrices it is not unique unless a certain additional assumption is introduced. If, for instance, the diagonal matrices of weights are constrained to be non-negative, three matrices are enough to have uniqueness in the rank two case of PARAFAC2. The authors are obliged to Richard Harshman for stimulating this research, and to the Associate Editor and reviewers for suggesting major improvements in the presentation.     

Consistency measures for pairwise comparison matrices

We propose new measures of consistency of additive and multiplicative pairwise comparison matrices. These measures, the relative consistency and relative error, are easy to compute and have clear and simple algebraic and geometric meaning, interpretation and properties. The correspondence between these measures in the additive and multiplicative cases reflects the same correspondence which underpins the algebraic structure of the problem and relates naturally to the corresponding optimization models and axiom systems. The relative consistency and relative error are related to one another by the theorem of Pythagoras through the decomposition of comparison matrices into their consistent and error components. One of the conclusions of our analysis is that inconsistency is not a sufficient reason for revision of judgements. © 1998 John Wiley & Sons, Ltd.     

A simplification of a result by zellini on the maximal rank of symmetric three-way arrays

Zellini (1979, Theorem 3.1) has shown how to decompose an arbitrary symmetric matrix of ordern ×n as a linear combination of 1/2n(n+1) fixed rank one matrices, thus constructing an explicit tensor basis for the set of symmetricn ×n matrices. Zellini's decomposition is based on properties of persymmetric matrices. In the present paper, a simplified tensor basis is given, by showing that a symmetric matrix can also be decomposed in terms of 1/2n(n+1) fixed binary matrices of rank one. The decomposition implies that ann ×n ×p array consisting ofp symmetricn ×n slabs has maximal rank 1/2n(n+1). Likewise, an unconstrained INDSCAL (symmetric CANDECOMP/PARAFAC) decomposition of such an array will yield a perfect fit in 1/2n(n+1) dimensions. When the fitting only pertains to the off-diagonal elements of the symmetric matrices, as is the case in a version of PARAFAC where communalities are involved, the maximal number of dimensions can be further reduced to 1/2n(n–1). However, when the saliences in INDSCAL are constrained to be nonnegative, the tensor basis result does not apply. In fact, it is shown that in this case the number of dimensions needed can be as large asp, the number of matrices analyzed.     

Independent feature processing in the visual system

Summary A recognition experiment with a set of simple visual patterns was performed. The patterns were composed of three binary features: Lines, angles and curve segments. The pattern sets differed with respect to the distance between components and the symmetry of the arrangement. The empirical confusion matrices were analysed assuming two different models: A model assuming independent feature analysis and a model that assumes that one of the features is analysed independently of the other two. The main result was that quite a large distance between features was necessary to achieve independent processing of the features. Moreover, in the asymmetric pattern set the predictions of the independence model were better than in the symmetric pattern set.     

Psychometric Qualities and Measurement Invariance of the Modified Self‐Rated Creativity Scale

The 13‐item self‐rated creativity scale (SRCS) initially developed for supervisory rating of employees’ creativity was modified by some researchers and used as a self‐report of creativity. However, it is not clear if the modified SRCS is psychometrically sound. The present study addressed this gap in three studies (N = 1,033). The exploratory factor analysis (Study 1) revealed a two‐factor solution after removing Item 9 due to low factor loading. Confirmatory factor analysis was then used in Study 2 to examine and compare the conceptual one‐factor model with 13 items (Model 1), one‐factor model with 12 items (Model 2), two‐factor model with 12 items (Model 3), and the 12‐item bifactor model with one general factor and two specific factors (Model 4). The results indicated that Model 4 is more superior to all the competing models. Study 3 further confirmed that the bifactor model, showed support to the reliability and convergent validity, and found partial metric invariance across Chinese and Malay undergraduates. Taken together, the modified (12‐item) SRCS is a psychometrically sound tool for self‐rated creativity in the Malaysian context.